The article introduces a new framework for feature selection in machine learning models. The approach uses a feature masking method, allowing the same model to be used during feature selection, eliminating the need for retraining. The framework also considers the importance of feature subsets as a whole rather than individual features. It shows significant performance improvements using LightGBM and Multi-Layer Perceptron as the ML models. The implementation code is shared to encourage further research and contributions.
Publication date: 23 Jan 2024
Project Page: https://arxiv.org/abs/2401.12644v1
Paper: https://arxiv.org/pdf/2401.12644
